internet traffic engineering
DESCRIPTION
TRANSCRIPT
Internet Traffic Engineering Using Multi-Protocol Label Switching (MPLS)
D.O. Awduche and B. Jabbari
Generalized Multi-Protocol Label Switching: An Overview of Signaling Enhancements and Recovery Techniques
A. Banerjee et. al.
Internet Traffic Engineering
Joachim Seilfaldet (joachse)Jonas Sæther Markussen (jonassm)
MULTIPROTOCOL LABEL SWITCHINGJonas Sæther Markussen
Agenda• Multi-Protocol Label Switching– Overview– Labels– Label assignment– Forward equivalence classes (FECs)– Label switched forwarding (LSP)– Control and data separation
• Generalized Multi-Protocol Label Switching• Traffic Engineering• Traffic Engineering using MPLS
Overview (1)• Multiprotocol Label Switching Architecture (RFC3031)• Overlay link network (OSI layer “2.5”)
– Eliminates the dependence on a specific link layer technology (e.g. ATM, SONET, etc.)
– Eliminates the need for multiple layer 2 networks to support multiple protocols• MPLS can carry many different kinds of traffic: IP, ATM, SONET,
Ethernet frames, etc.
• Constraint-based forwarding– (Optional) assignment of labels prefixing packet headers– Forwarding no longer constricted to packet destination
only
Overview (2)• Introduces connection-oriented routing to
legacy IP routers– Does this by establishing “virtual links” using label
switched paths (LSPs)• Provides means for traffic engineering (TE)– Can manipulate link-state advertisements (LSAs)– An easy-to-maintain virtual topology– With TE, MPLS can establish alternative paths to
avoid congested areas• Enables fault tolerance if a link goes down
Labels• MPLS introduces labels
– Originally called “tags” in earlier works by Cisco and others– Fixed-size header
• 20-bit Label Value• 3-bit Traffic Class (QoS priority and ECN)• 1-bit Bottom-of-Stack flag• 8-bit Time-to-Live
– Unlike ATM and frame relay, labels can be stacked• Allowing hierarchical arrangement of frames
– Labels are prefixed to IP headers and to each other• Enables fast look-ups (more on this later)
Label Value Class BoS TTL
Label assignment (1)• Packets enter a MPLS domain through an ingress
node and leaves through an egress node– These are typically label edge routers (LERs)
• Ingress nodes assigns (pushes) to and egress nodes removes (pops) labels from packets coming in
Exit node
Entry node
LER
LER
MPLS domain
Label assignment (2)• Three label operations:
– Push (impose, assign)• Encapsulates the packet in a new MPLS layer• Allows hierarchical routing
– Used by e.g. MPLS VPN (L3VPN)
– Pop (dispose, remove)• Remove uppermost label• When the last label is popped, the packet “leaves” the MPLS
tunnel/domain• Usually done by the egress router (exit node)
– Can be done by the preceding LSR for offloading the egress router penultimate hop popping (PHP)
– Swap• Simply replaces the label and forwards packet along the path associated
with the new label
Forward equivalence classes• Label assignment is based on forward equivalence classes
(FECs)– Packets belonging to the same FEC has the same labels– FECs can be defined differently:
• Based on enter (ingress) nodes and exit (egress) nodes• Based on service class, requiring similar QoS or packet treatment
across the MPLS domain• Packets belonging to the same flow• Combinations of those above• FECs are associated according to some policy formulation
• Packets belonging to the same FEC traverse through the same path (or multi-paths)– This is called a label switched path (LSP)
Label switched forwarding (1)• “Virtual links” presented to above layers in the OSI stack
– These are called label switched paths (LSPs)– From one label edge router (LER) to another
• Typically the same as ingress and egress nodes
– Established (and tore down) by a signaling protocol (more on this later)– This introduces connection-orientation in networks that originally were based on packet
switching (PS) unified data carrying for both PS and circuit-based• Routers in the MPLS domain that forward both labeled packets and conventional
IP packets are called label switching routers (LSRs)– Label edge routers (LERs) are usually LSRs with label stacking functionality
LERRouter
LSR
Router
LER
LSR
Phys. links
LSP paths
Link from IP perspective
Label switched forwarding (2)• LSP update policy can vary:– Predefined (strategic)
• Careful planning of the virtual topology• Considerations and forecasting to traffic patterns• How, when and where to activate new LSPs to address
performance issues in the network
– Ad-hoc (tactical)• Establishment and managing of LSPs to divert traffic away
from congested network resources to under-utilized alternatives
• A “hybrid approach”: LSPs control traffic parts in some segments of network while interior gateway routing protocol metrics are used in other
Control and data separation (1)• MPLS functionality is separated into two “planes” with different purposes• The planes are decoupled and independent
– Clear separation of the control plane from the data plane in network switching elements
– Even further separation in Generalized MPLS (GMPLS)
Control
Routing Path selection
SignalingLocal
resource control
TransportLabel swapping
Packet forwardingPacket treatment
From the article, Fig. 3
Control
Routing Path selection
SignalingLocal
resource control
TransportLabel swapping
Packet forwardingPacket treatment
Protocol Transactions
Bearer Channels
Control and data separation (2)• Control plane– Control protocols are software processes that
communicate across node boundaries– Distribute and manage:
• Network topology• Resource availability
– Establish and tear down LSPs• Signaling protocol
– Label distribution protocol (LDP) for best-effort hop-to-hop paths
– RSVP-TE (or CR-LDP) for traffic engineering purposes and end-to-end virtual circuits
Control and data separation (3)• Forwarding plane• Label swapping operations
– Look-up tables
• Packet treatment functions– Scheduling– Queue management– Rate shaping– Policing
• Usually implemented in hardware– High speed operations
GENERALIZED MULTI-PROTOCOL LABEL SWITCHING
Joachim Seilfaldet
Agenda• Multi-Protocol Label Switching• Generalized Multi-Protocol Label Switching
– What is GMPLS?– Implemented interfaces to support– Enhancements to Signaling– Hierarchical LSP Setup– GMPLS Protection and Restoration Techniques– Path Switching– Line Switching– Protection Mechanisms– Restoration Mechanisms
• Traffic Engineering• Traffic Engineering using MPLS
What is GMPLS?
Generalized Multi-Protocol Label Switching• Next generation implementation of Multi-Protocol Label Switching• Extends to support a wide range of LSP for different network devices.• Extensions made to IP router protocols (OSPF and IS-IS)• New Link Management Protocol
Multi-Protocol Label Switching Recap• Works as an extension of IP• Control plane is logically separated from data plane. • Referred as a “Layer 2.5” protocol. Layer 2 (Data Link Layer) and Layer
3 (Network Layer).
What is GMPLS?
Control plane concepts can be used in other switched transport technologies
• Packet Switched Networks
• A label represent a short tag attached to packet
• Time-Division Multiplexing Networks
• A label represent a time slot
• Wavelength-Switched Networks
• A label represent a wavelength
• Fiber-switched Networks
• A label represent a fiber
Implemented interfaces to support
• Packet Switch Capable Interfaces (PSC) If a node recives data over this interface, it will be able to switch the
recived data on a packet-by-packet basis based on the label attached.
• Time-Division Multiplexing (TDM) Will be able to multiplex or de-multiplex channels within an payload.
• Lambda Switch Capable Interface (LSC) Will be able to recognize and switch individual lambdas within the
interface.
• Fiber Switch Capable Interface (FSC)Will be able to switch the entire contents to another interface (without distinguishing lambdas, channels or packets), such as optical cross-connects (OXCs) .
Enhancements to Signaling
• GMPLS require LSP start and end on similar device • For example, SONET TDM.
• Necessitates a separate control plane transport network.• GMPLS is extended to allow control plane to be physically
diverse from the associated data plan.
•Enhancements have been made to the label distribution protocol RSVP-TE to support GMPLS.
Hierarchical LSP Setup
• Occurs when a new LSP is tunneled inside an existing higher-order LSP
• Serves as a link through other LSP• Nodes at border of regions are responsible for
forming higher-order LSP and aggregating lower-order LSPs.
Hierarchical LSP Setup
Figure shows how hierarchical LSP setup is performed over different types of network types.
Hierarchical LSP SetupR0 R1 S2 O3 P4 P5 P6 O7 S8 R9 R10
Timeline
Path 1Path 2
Path 3
Path 4
Resv 4
Resv 3
Resv 2
Resv 1
LSP4 completes
LSP3 completes
LSP2 completes
LSP1 completes
GMPLS Protection and Restoration Techniques
Protection and restoration is addressed using two techniques Path Switching Line Switching
Fault management consist of Detection Localization Notification Mitigation (Done with protection and restoration)
Protection Mechanisms
• Efficient use of protection requires• Distribution of relevant link properties
Protection bandwidth Protection capabilities
• Establish secondary paths through network• Signal switch from primary path to backup
Path Switching
•Failure is addressed at path endpoints.
•Path protection Protection path is pre-allocated. Resources for protection path is reserved,
specifically to handle traffic from path that is protected.
Path restoration Restoration of path needs to happen “on-the-fly”
or to be pre-computed and cached at endpoints. No resources are reserved in case of a failure.
Line Switching
•Failure is addressed at transit node, where failure is detected.
•Span Protection Traffic is switched to an alternate parallel channel
or link connecting same two nodes.•Line Restoration Traffic is switched to an alternate route between
two failing nodes. Passing through additional intermediate nodes.
Protection Mechanisms
•1+1 protection Data transmitted simultaneously over two
paths. Will receive on backup path, in case of errors
on working path.
M:N protection M pre-allocated backup paths shared between
N primary path.
Protection Mechanisms
Note: 1:N and 1:1 are just special cases of M:N
1:N protection 1 pre-allocated backup path shared
between N primary paths
1:1 protection 1 dedicated backup path is assigned for each
primary path
1+1 span protection
Transmitted simultaneously over two disjoint channels
Receiver discards packets from protection path
On failure in working path will switch to protection path
Protection Path
Working PathA B
1:1 span protection
• Transmitted only over primary channel
• Backup channel has been computed
1. Link Management Protocol will localize failure.
2. RSVP refresh message will indicate a path switchover.
3. Both nodes make switch to backup channel.
Backup path
Working PathA B
(1)
(2)
(3) (3)
Restoration Mechanisms
• Designed to..• React to failures quickly• Use bandwidth efficiently
Slower than protection mechanisms• Dynamic resource establishment• Route calculation
Restoration Mechanisms
•Path restoration• Optimization can be done to
speed up process.• Pre-computed paths and
cached at head and end nodes.
• May reuse nodes in original path. BOOM
Restoration Mechanisms
•Line restoration• Beneficial for connections
that span multiple hops• May brake TE
requirements • Constraints must be
forwarded, for intermediate nodes to be able to do line restoration
BOOM
TRAFFIC ENGINEERINGJonas Sæther Markussen
Agenda• Multi-Protocol Label Switching• Generalized Multi-Protocol Label Switching• Traffic Engineering– Limitations of legacy IP networks– Traffic engineering in general– Traffic engineering process– Overlay traffic engineering
• Traffic Engineering using MPLS
Limitations of legacy IP networks (1) • Routing– Conventional shortest path routing protocols
• Packet-switching• Usually link-state (OSPF or IS-IS) or distance-vector• Simple and distributed• Link layer dependant
– May even be so crude as 1:1 mapping of physical links!
– Routing based on simple hop-to-hop metrics• Mainly calculated from bandwidth
• “Best effort” environment– Initially, this was why it was so successful– Not reliable with today’s QoS and performance demands
Limitations of legacy IP networks (2) • Poor resource allocation– Under/over-utilized paths due to shortest paths
algorithms using link state metrics (usually bandwidth) as the only link weight
– May result in congestion even when excess capacity exists in alternative paths!
• Virtually no traffic measurement methods– Absence of reliable data• Lack of ability to produce traffic matrix
Traffic engineering in general (1)• Aims to improve the unreliable and limited
behavior of IP networks– Link-metric based shortest path route computation
• Distributed shortest path first algorithms, e.g. Dijkstra’s
– Resource availability and traffic characteristics are not taken into considerations when routing traffic
– Not feasible to estimate traffic matrices from router interface statistics due to distributed nature of IP
– When congestion occurs, hard to determine which source-destination pairs contributes
Traffic engineering in general (2)• Goal is to address issues concerning:
• Traffic control• Resource control• Measurements
• Different types of traffic engineering methodologies and TE classifications
Dynamic vs. Static Online vs. Offline Predictive vs. Descriptive
Proactive vs. Reactive Time Dependent vs. State Dependent
Open Loop vs. Closed Loop
Tactical vs. Strategic Local info vs. Global info
Centralized vs. Distributed
From the article, Fig. 5
Traffic engineering process• Traffic engineering is an continuous process
– Policy formulation• Guidelines for traffic management, traffic control and operation of the network
– Data acquisition• Empirical statistics are gathered through measurement
– Traffic patterns, link utilization, traffic trends, packet drop statistics– Mathematical models can be used where statistics are unavailable and/or in supplement
– Analysis and characterization• Based on the workload derived from the measurement phase
– Performance optimization• Continual and iterative process• Traffic control: Manage inflow to the network and mapping of traffic to network resources• Altering network topology: Adding links, increase or decrease link capacity, etc.• Controlling local packet treatment: Queuing, scheduling, dropping policy, etc.
Policy Formulation Data Acquisition Analysis & Char. Performance Opt.
Traffic engineering work cycle
From the article, Fig. 4, simplified
Overlay traffic engineering (1)• Early works revealed that virtual connection-
based abstractions with originating connection control compensated for legacy IP routing issues in dense topologies
• ISPs introduced virtual circuit (VC) switching technologies, i.e. ATM and frame relay, into IP infrastructure
Overlay traffic engineering (2)• VC introduced with an overlay configuration– Elements of the VC technology are placed at the core and are
surrounded by regular IP routers– VCs serve as point-to-point connections between routers,
which routing protocols establish adjacencies routers connected by a VC appears as neighbors in the IP routing layer
ATM switch
ATM switch
ATM switch
ATM switch
IP Router
IP Router
IP Router
ATM network
Physical links
Links as seen from IP perspective
Overlay traffic engineering (3)• Many advantages of an overlay structure– Decoupling of control planes for the virtual-circuit-
based network and control plane of the IP network• Can use conventional IETF IP protocols (OSPF, BGP, etc)
– Virtual circuits can be rerouted to move traffic away from congested resources onto under-utilized alternatives
– Allows the service provider to derive estimates for a traffic matrix by monitoring traffic flow over virtual circuits
Overlay traffic engineering (4)• Disadvantages with IP over ATM and IP over
frame relay– Added cost of building and managing two
independent networks with dissimilar technologies and different semantics
– The so-called O(N2) scaling problem• The number of VCs grows as a function of the square of
the number of routers in the network• …and so does the number of adjacencies between
routers
TRAFFIC ENGINEERING USING MPLSJonas Sæther Markussen
Agenda• Multi-Protocol Label Switching• Generalized Multi-Protocol Label Switching• Traffic Engineering• Traffic Engineering using MPLS– Comparison to the overlay model– Protocol extensions– LSP-tunnels– Traffic engineering using MPLS
Comparison to the overlay model• MPLS introduces constraint-based routing, which makes it very
useful for traffic engineering (TE)• Provides an overlay model in an integrated fashion on a single
network element• Advantages of MPLS for TE relative to the overlay model
– Fewer network elements– Lower operating costs– Greater reliability due to fewer network elements exist along the
routed path– Potentially less latency– Simplified network architectures
• MPLS also supports the overlay model, giving service providers the option to deploy overlay or integrated solutions
Protocol extensions• Requirements to MPLS in IETF RFC-2702
– Effective means for MPLS to deploy and implement various TE policies• Resulted in extension of legacy IP routing protocols and signaling
protocols• BGP (version 4, RFC4271)
– ISIS-TE, OSPF-TE (RFC-3630)• Extended to advertise new types of capabilities and constraints associated with
links
– RSVP-TE (RFC-3209, RFC-5151)• Earlier CR-LDP was used, but was deprecated (Feb. 2003) and replaced by RSVP-TE• New objects added to RSVP to support establishment & teardown of LSPs w/
behavioral attributes• Can establish parameterized explicit LSPs and assign network resources to them
• The extensions make out the MPLS-TE control plane– Requirements expanded to encompass capabilities to support Diffserv-
aware traffic engineering
LSP-tunnels• “Traffic trunks”
– Traffic belonging to the same class that are routed through a common path or multipath (LSP-tunnel)
– “LSP-tunnel” refer to both the “traffic trunk” and to the LSP it traverses• TE extensions to MPLS support assignment of attributes to LSP-
tunnels– Bandwidth characteristics, resource affinities, resilience attributes,
priority attributes, preemptive capabilities, with more• Simplified establishment of LSP-tunnels
– Establishment is done by configuring endpoints plus desired performance and behavioral attributes at an originating LSR
– The LSR will employ constraint-based path computation algorithm to compute a path through the network satisfying the LSP-tunnel specifications subject to various constraints that exists within the network
Diffserv and MPLS (1)• Two important components of resource
allocation in IP networks– MPLS: Global resource allocation within a given
domain, constraint-based routing with bandwidth resource allocation
– Diffserv: Local resource allocation, “per hop behaviors” (PHB) buffer and link resources to packets based on the Diffserv code point (DSCP) in the packet headers
Diffserv and MPLS (2)• MPLS has basic support for Diffserv– Diffserv behavior aggregates can be mapped onto
LSPs• Two types of LSPs support this capability, EXP-inferred-
LSPs (E-LSPs) and Label-inferred-LSPs (L-LSPs)
– MPLS support Diffserv aware traffic engineering• Derives from the fact that original MPLS-TE proposals
focused on the optimization of aggregated traffic trunks, not taking to consideration the issue of preferential treatment to different types of traffic in a Diffserv environment
Traffic engineering using MPLS (1)• Considerations– Global/prevailing network constraints– LSR interface attributes
• Local packet treatment
– LSP parameters and LSP paths from originating LSRs• Strategic (predefined) vs. tactical (ad-hoc)
– LSP topology• Maintainability vs. loss of efficiency• Large vs. small number of LSP-tunnels
– Load balancing• Multiple parallel LSPs with common endpoints• Dynamic vs. static, open loop vs. closed loop etc
Traffic engineering using MPLS (2)• Network survivability– MPLS offers enhanced survivability capabilities
• Different types of protection, restoration and local repair schemes
– Backup LSP-tunnels and explicit LSP routes• Measurement considerations– Monitor
• Routes traversed by each LSP in the network• Bandwidth requirements of each LSP• Dynamics of LSPs in the network
– In Diffserv environments, it is desirable to measure the dealy along an LSP under different conditions